Notes: I merged the adapter back with Unsloth. Looks like you should use the Mistral instruct format.
L quants (or more), for fun/testing, probably prefer bartowski or mradermacher's quants if available
Q2_K_L, Q4_K_L, Q5_K_L, Q6_K_L, are using Q_8 output tensors and token embeddings. imatrix is done using bartowski's imatrix dataset
